203 | The talent war goes nuclear: Meta Launches Superintelligence Labs, “Fair use” courtroom win, Microsoft axes 9,000 jobs and more AI news for the week ending July 4, 2025

23:30And so if you are in a leadership position in a small or medium business, and you understand what is happening, and you are terrified in it, you have bigger fish to fry every day and fires to take care of, and you don't have the bandwidth or capacity to deal with AI training, please reach out to me on LinkedIn or find the link in the show notes of this podcast on a way to book a meeting with me and to discuss how to provide the best possible training for your employees in order to start moving forward with the AI adoption or accelerate your existing adoption. And if you're an individual and your company is not providing that, come and join our AI Business Transformation course.

24:02We have helped thousands of business people, mostly business leaders and entrepreneurs, in learning how to apply AI most effectively across multiple aspects of the business. The next course starts on August 11th. It's four weeks, two hours a week, which means within one month, you'll be in a completely different situation where you are right now in your AI knowledge and skills because it's a lot of hands-on experimentation across multiple aspects of business from data analysis to content creation to sales automation and so on. So come and join us. Don't wait because while last year this was nice to have, it's becoming essential for employees and for companies to have this.

24:35And if you are in a small or medium business, it gives you a huge opportunity to move ahead faster than your competitors, at least based on this recent survey. And the last topic on the jobs aspect is that Anthropic just launched the Economic Future program, and they announced it on June 27th. And their goal is to help study of AI economic impacts and develop policy solutions and help develop policy solutions to assist us in dealing with job displacements created by AI. If you remember, I was sharing with you recently that Dario Amadei believed that AI could eliminate 50 % of entry-level white-collar jobs in the next few years, with overall unemployment rising to 20 % within one to five years.

25:11That's one to five, meaning it could start next year. And their new initiative offers up to$50 ,000 in grants for empirical research of AI labor market effects and Claude API credits for partnerships with research institutions to be able to help them do the research with AI, which kind of sounds counterintuitive. Let's use AI in order to find how AI is going to impact future jobs, kind of like fighting fire with fire. And this is another great example of how Anthropik are doing things a little differently and really trying to stay behind their core values of finding better, safe ways in order to introduce AI.

25:44I said many times before that all I see with those leaders is that they come and saying, hey, look, you've got to be careful. We've got to do this. We've got to do that. But they're not actually offering any solutions. And so kudos to Dario Amadei and Anthropik for actually doing something proactive, at least on the research side, that hopefully help us move in the right direction. And I really hope that other organizations, including government agencies and other companies from the AI space, will move in a similar direction and put enough money and resources in order to help us all figure this out faster.

26:11I'm personally doing whatever I can in providing training, in providing free webinars, in doing this podcast, in doing training sessions, and a lot of other things that are doing in order to try to help people deal with the situation. But I think we need something much, much bigger from a country perspective and from a global perspective in order to figure this out properly. But since we mentioned Anthropic, this is the perfect segue for the last really important topic of this week. A federal judge ruled on June 24th that Anthropic use of copyrighted books to train its AI model Claude is fair use.

26:39It is the first time that a judge endorses this defense by the AI companies against all the lawsuits that were filed against them. So let's go back a few steps and do a little bit of background. All of these AI companies has trained on content they do not own. They basically scraped every piece of information they can from the internet and used it to train their models. When they got sued for doing that, they claimed fair use. They basically said, this is just like humans are learning. The fact that the AI is reading the website or the blog or the book or whatever the situation is, is just like people doing it.

27:10It's just doing it more efficiently. And it's not actually reproducing the same content. It is using it as part of its knowledge in order to create new stuff. And the judge, William Alsop, and I hope I'm not butchering his name, called Anthropics training process, and I'm quoting, exceedingly transformative, basically saying that the content that it's generating is not replacing or copying the original work, but it is dramatically different. And hence, it is aligning with fair use under the US copyright law. That being said, Anthropic is on trial for downloading over 7 million pirated books to build its central library of data.

27:44And the same judge ruled that this is not covered by fair use. So where does that leave us from AI world as a whole. First of all, I do believe, like I said many times before in this podcast, this will end up with the Supreme Court. Now with the Supreme Court and the current government that we have, I do think it will lean towards this direction. I obviously don't know that for a fact, but that's what I think that will happen. But what he made very clear is that you cannot steal the information in order to train, meaning getting pirated books, not okay. Buying books in order to train your model is okay.

28:13And probably the same thing is true for many other types of content that these companies are put their hands on, whether lawfully or not. Now, in the beginning, Anthropic actually purchased millions of physical books that sadly they destroyed after that instead of donating in order to build their initial research library. And the judge claims that this is fair use, meaning if you're buying, if you have legal access to the material, then you can use it in order to train your models. So while this is definitely not the final aspect of this, this is a very important precedence moving in the right direction of two different things.

28:44A, it's saying what you can and cannot do. And B, it will probably push forward a lot more licensing deals that we've seen in the past year in order to allow these companies to train with data that they actually own and deliver us better and better results without breaking the law. What will happen with all the stuff that they already did that was not legal? It's very, very hard to say. I assume there's going to be some kind of a compensation package with billions of dollars changing hands, lots of lawyers making a lot of money, and the actual creators of the content making a little bit of money, maybe, but I don't see how this is going to stop the AI revolution.

29:17If you remember, we spoke in the past two weeks about the lawsuit by Disney and Universal against Midjourney, and we said that setting up a precedence against a small company will have a trickle effect to the rest of the industry. This is exactly the same thing the other way around, which is why this is so important. Again, it's the first time that somebody in the legal system has said that training models on existing material, as long as you had legal access to it, is fair use. And now to some rapid fire items and we're going to stay in the legal arena because there's been a very interesting article in Reuters this week has to do with how law firms are using AI and how it's affecting the legal world as a whole.

29:52So let's go back a little. In 2023, an Avianca flight injury case, the lawyers cited six fake cases generated by ChatGPT because they did not know it can make stuff up. That led to sanctions against the actual lawyers in the land bank ruling on AI legal risk, highlighting the need to actually verify the information. Well, that was 2023. People did not know better. Well, the same exact thing happened this year for a lawyer from Morgan & Morgan, where he cited several cases that don't exist because he did the research with AI. So this is the risk. But on the other hand, AI tools like Harvey and CoCounsel currently slash document review by up to 80 % because of its ability to analyze thousands of pages in minutes or a few hours versus days and weeks, and that is freeing lawyers and paralegals to do more important work than just reviewing documents.

30:39Now, while this, again, sounds great, I think it actually generates a very serious risk to the legal world as it is today, and to be fair, more than just the legal world, because this automation could dramatically disrupt the way these companies work and make money right now. All law firms, or at least most law firms, make most of their money based on billable hours, and a lot of these billable hours come from exactly this grunt work of reviewing information, summarizing information, searching for information, and combining that information into making a case. And if an AI can do that in minutes instead of days, all those billable hours go away.

31:13And while some legal work definitely still require lawyers, especially when it comes to dealing with humans who are involved with these legal situations, a lot of the hours may go away, which may have a significant impact on law firms' revenue that might be significant enough to take the company down. And while right now it might be a benefit to law firms because they can do the work a lot faster and probably still charge roughly the same amount of money as they did before, customers will learn what AI is capable of doing and hence will push the law firms to dramatically reduce their prices. There's already, in a recent survey, already law firms said that 26 % of companies are feeling client pressure to adopt AI in order to reduce the cost of the legal support that they're getting.

31:49In general, I think the hourly rate model of consulting, legal, and other industries as well is at a very, very high risk because it will become obvious to everyone that the amount of hours required to do things is going to be cut not by 5 % to 10%, but by 95 % in some cases, which means if what you're doing is based on hourly rates and in an area where AI can do the work better, you are at a very, very serious risk. And from the legal world to the classroom, a Gallup poll reveals that over 25 % of K-12 teachers already use AI in the learning system. This is a huge spike from close to zero about a year ago, but at the same time, 80 % of educators report that their districts lack policies for responsible AI use or any kind of AI use, which is creating basically the Wild West, where every teacher can do whatever they want in the classroom with or without AI.

32:35And this is why the American Psychologist Association just issued a warning that saying that AI may impair adolescents' ability to distinguish between humans and simulated empathy. So this goes way beyond learning. We're talking about the emotional development of kids because they will be dependent on AI communication for many things that they're doing. There's been a lot of research in the past few years, even before ChatGPT, rolling back all the way to 2021, that is showing AI's potential for tailored personalized instructions that lead to much better results, including some actual large-scale tests that were done in Africa through providing personalized training with AI that was able to do in weeks the amount of knowledge acquisition in individuals that usually took months.

33:15And so I do think that this is the biggest opportunity the education world had in decades. President Donald Trump, in his April executive order, emphasize AI literacy in K-12 as an important aspect of what they want to do, but nothing has been done since. It's only been two months, to be fair, but I do think that this is a huge opportunity for teachers to become mentors and really help young individuals grow while allowing AI to do the tedious aspect of training in a way that will be perfectly aligned with the specific needs of every kid at the specific time that they need it, 24-7, 365, in the right ways that helps the student learn, whether it's a game, a video, a reading segment, a test, whatever is the thing that will help every particular kid learn in the most effective way that particular part of the material.

33:58But our education system is really big and really slow to react. But I truly hope we'll start seeing more and more of this happening because as I mentioned, I see this as a huge opportunity. Now, we cannot have a week of news without stuff about OpenAI. So OpenAI just signed a$30 billion a year deal with Oracle to lease 4.5 gigawatts of computing power, which is one of the largest cloud deals in history, and it is a part of the Stargate initiative. So to take you back, the Stargate initiative is a$500 billion joint venture between Oracle, SoftBank, and Abu Dhabi's MGX fund, together with OpenAI to build data centers all around the world.

34:34And this is just one deal out of all the deals they're going to sign over time. So it's not new because it's 30 billion out of 500 billion, but it's 30 billion that are actually materializing in the near future. Just to explain how much is 4.5 gigawatts of capacity, it's equivalent to the amount of power that is generated by four nuclear reactors. So that's a lot of power that drives a lot of computers that OpenAI needs in order to continue to drive its crazy growth. But going back to the type of jobs that it generates in the near future, this deal is starting at fiscal year 28, and it's going to more than double Oracle's current$28 billion cloud infrastructure revenue.

35:10So within the next few years, they will have to build this capacity. They will double the capacity of the compute that they have right now, which means they will need electricians and computer hardware people and construction people in order to build and connect and operate all of these data centers, which means there's going to be a lot of these kind of jobs compared to other jobs that will become harder and harder to get. Another big news from OpenAI is OpenAI just announced that they are starting to provide consulting, offering in-house engineers to customize AI models for enterprise and government agencies.

35:40That basically means that you can bring OpenAI's employees to work within your organization in order to fine-tune models and have the best setup for your particular organization, there's one small caveat, which is you have to be willing to spend at least$10 million a year in order to get these services, going back to the growing gap between enterprise and large organizations' ability to implement AI compared with small to medium businesses. Now, this is OpenAI going into yet another field of another company, as they've been doing all along, going into more and more fields and getting their hands into more and more buckets of revenue.

36:11So if you think about companies like Palantir or companies like the Big Four and Accenture, with the personalized hands-on consulting that they're providing, OpenAI is now starting to provide that on its own. The benefits are very obvious, right? If you're in a large organization and you can have the people who are developing the model and know exactly how to work with it and train it, help you do it for your needs, it can dramatically accelerate the implementation and increase the chances of success, which makes a lot of sense, again, as long as you're willing to spend at least$10 million a year.

36:37So if you cannot spend$10 million a year, what can you do as a smaller business? I mentioned that before, training and education. Get your people to a level where they can do at least small things and work on strategy in your company on how to do bigger things without spending$10 million on OpenAI Consulting. And the last piece of news about OpenAI this week has been the soap opera around the I.O. trademark, which has been going back and forth on X between Sam Altman and a company that nobody heard of until about two weeks ago, a company called I.O. that is spelled really weird. So there's I.O.

37:07I.Y.O. and I.O., the letters I.O. So the story goes like this. the company IYO has been developing since 2021 a hardware solution that will be able to be used in order to engage with AI. In their particular case, it's a wearable earbuds that you can use in order to talk to the AI in a seamless way. They've been engaged with OpenAI since 2022, trying to get OpenAI to invest in them or maybe even buy them out. And these discussions have been going on for several years, including until the spring of 2025, when OpenAI announced they're bringing in Johnny Ives' company called IO that is going to do, yes, you guessed correctly, develop hardware devices that will allow humans to engage with AI.

37:49So on one hand, you have a company that has exposed everything that it's doing to a potential investor. And on the other hand, you have the same exact investor come out with a competing product a few years later that has a very similar name and that sounds exactly the same. So for now, the judge granted IO the company a temporary restraining order that is barring OpenAI from using the IO brand, at least until the October hearing. So what happened so far is OpenAI removed the IO mentioned from everything, including its website and all of its promotional materials. Now, Sam Altman's response, what he called the lawsuit silly, and yet IO's general counsel told Fortune, we're not surprised by Mr.

38:27Altman's dismissive tone, but federal judges don't issue restraining orders based on CV claims. So what is happening here is I don't think that it will stop OpenAI for developing their products. I'll share with you in a minute what we've actually learned because of this case, but it may force them not to use the brand name IO for anything related to what they are doing right now. What is very, very clear that IO, the company, IYO, is getting crazy exposure because it's a company nobody heard about until this past week. And now everybody on X is talking about them and about the current situation.

38:57So if they got something out of it, it's a lot of good exposure to what they're doing. Hopefully they also have a good product, but I don't know that for a fact. But what this case allowed us to do is to learn a little bit about what Johnny Ive and Sam Altman are planning, because the court filings for that dispute is confirming that the device from OpenAI and Johnny Ive is not going to be a wearable or an in-ear gadget, which removes glasses and smartwatches from the list, right? Because these are wearables and it is designed to be a pocket-sized third-core device. This is from the actual court filing itself.

39:29So those of of you who don't know the story, Johnny Ive is the iconic Apple designer that is behind some of the most iconic devices in history, including the iPhone and the iMac and many other devices that Apple generated through the years that completely changed the way we engaged with computers. He's now working with Sam Altman to develop the next generation of how humans engage with computers, in this particular case, AI. Now, apparently, Johnny Ive really likes pens, and he has a collection of some vintage, really expensive pens. So that might be a hint of what they might be developing. He's also the guy that designed the pen for the iPad.

40:01That is very successful that a lot of people love using. So that might be another hint. Does that mean that that's what it's going to be? No. But the core documents also indicate that the device won't ship until at least 2026 with mass production coming in 2027. So it will be a while, I think, before we actually learn what they are doing. And also this week, OpenAI shared a very interesting article on their involvement with GenSpark. So GenSpark is one of the general agent tools that allow you to build almost anything you want with AI agents from presentations to research to websites to basically anything you ask it to do.

40:32It's a very powerful tool. I love using it. If you go back to episode 196, you will find an episode called How to Safely Run Powerful Agents Like Manus and GenSpark, where I shared how to use tools like this in a safe way without them having access to any of your data and or your computer. But the interesting thing in this piece of news is that GenSpark, without investing any money in paid marketing, just hit$36 million in annual recurring revenue, and they are using OpenAI's multi-modal models and real-time API in the background. And this information was released not by GenSpark, but by OpenAI themselves.

41:02So OpenAI has shared that GenSpark super agents enables users to automate tasks like making phone calls and creating presentations, generating videos and simple websites, all with no coding experience and no knowledge in technology. And what they shared is that in the backend, it is actually running GPT 4.1 for research and the GPT image generator for visuals. That's not the only stuff that they're integrating. They're also integrated with nine specialized large language models and over 80 tools that it's connected to in order to do the things that it's doing. But the core behind it is GPT-4-1 with one million token context window to handle long documents and the long operation that it needs to do.

41:36Now, going back to OpenAI's transformative nature in partnering with them, Eric Jing, the CEO of GenSpark, said, and I'm quoting, their APIs didn't just power our models. They helped our 20-person team build, launch, and scale faster than anyone thought possible. So going back to if you have the money and you have the connections, you can have OpenAI help you build products that can scale at crazy speeds. Now, what does it show? It shows two things in my mind, and both are important. One, that if you have the right idea and a small team of talented people and access to the right AI APIs behind the scenes right now, you can build extremely successful companies extremely fast.

42:11There are several examples of that right now, and GenSpark are not the only ones. But the other thing that is not shared in this piece of information by OpenAI is the big question, which is what happens to GenSpark once OpenAI generates the same exact solution on their own and makes a super agent solution part of ChatGPT, which there is zero doubt in my mind that they are planning. There's very little doubt in my mind that they already have something like this that they're just not releasing because it's not ready yet. It's not safe enough. It's colliding with something else that they're doing, but that's the direction that they're going.

42:40It is very, very obvious to me that this will be coming. And hence, what will happen to GenSpark as a company once OpenAI themselves has a tool that competes with GenSpark? At that point, they're probably done because OpenAI is a known brand. People will trust the name. People will trust the data security. And the chances that people are going to use OpenAI significantly exceeds the chances that they're going to use GenSpark to do exactly the same task. Not to mention the fact that people are already using OpenAI and ChatGPT. So they already have it on their computer. It's just going to be another feature, most likely a paid feature.

43:09I don't know how much it's going to cost, but it's not necessarily going to cost more than also getting access to GenSpark. So by paying the same amount of money, you will keep on using one tool for all the things that you've used before, plus the GenSpar capability. This is not going to be the first startup or the first industry that a new feature within Chachipiti and or Anthropic and or Gemini is going to completely eliminate. And this is one of the biggest risks that I see moving forward. It is very, very clear that these companies are going after everything they can. They raise enough money to do that and they're going to come up with everything people need, which is every main feature that either business or individuals need will be supplied by these companies at a certain point.

43:47And if you are developing a startup company that is doing a generic something, your chances of not getting wiped out by a feature in Chachipiti or Gemini or Cloud are relatively small. So if you want to develop AI-based solutions, you have to think about something that tailors more things and more aspects than just being an overlay that provides an easier usage or some additional basic capabilities on top of the AI tools. It needs to be well-integrated and well-customized for something very specific. And with that in mind, I want to give you a few additional examples, all relevant from this past week.

44:19So the AI voice startup Eleven Labs, which a lot of people love using, are working towards an IPO. They have raised a significant amount of money, and their CEO just confirmed that in an interview with CNBC. The company plans new offices in Paris, Singapore, Brazil, and Mexico, and that's complementing their existing hubs in London, New York, Warsaw, San Francisco, Japan, India, and Bangalore. So they are building more and more offices around the world, doing a lot of partnerships with many different companies. And while they've been highly successful, and I think they really have a great product, I don't know how much chance they have once these things, once the same capabilities that 11Lab provides will become a standard included feature in things like ChatGPT, Gemini, Claude, etc.

45:01And yes, they've been first movers into that field. And they're probably the most known name right now as an AI voice generator, but they will have significant competition from above with companies like Chachapiti and from below with a company like Nari Labs, which was established by two college students that is generating a voice tool called Dia that has incredible voice capabilities. I shared that with you in the past. And so that's another company that I think has done extremely well that I don't see a very solid future for just because I don't think the business model is there for them to survive.

45:30Smaller companies will be able to do the same for almost free, and bigger companies will give that as a feature for free. And so what kind of business they can find in between, their only chance is to build more and more integrations into existing niche universes that will make it hard to replace them afterwards in a way that will make them embedded into several different environments, which will be harder to replace. Another very large company that is facing a similar situation where trying to combat ChatGPT is Microsoft. So apparently, ChatGPT keeps on growing in the business world at a very fast pace.

46:02They already have over 3 million paying users, while Microsoft Copilot user base has grown to 20 million, but then has stagnated at that number for a while now, with more and more companies are switching from Copilot to ChatGPT, and more and more companies that want to start with AI start with ChatGPT without even trying Microsoft Copilot. And that is without even trying Microsoft Copilot, and that's based on a report on Bloomberg. Microsoft sales team in internal memos are saying that they're seeing significant challenges in differentiating Copilot from ChatGPT. Well, many customers like ChatGPT because they're saying, A, it's better, which I agree 100%, and B, they already know it because they use it at home for other things, so they feel more comfortable with the user interface and with how to engage with it, and hence they prefer ChatGPT over Copilot.

46:44So think about it. There are over 800 million weekly active users on ChatGPT and 3 million business users. That leaves us with 797 million people who use ChatGPT not for work. And guess what will be their preferred tool to use when they do, when they can't use it at work? It's going to most likely be ChatGPT. And so that generates a serious risk, not just for small businesses, but also for a giant like Microsoft. But based on Microsoft, the biggest problem is people do not know how to properly prompt and use Microsoft Copilot. And so they launched Copilot Academy to try to address this. I don't think that's enough.

47:15I think the right way to do this, and I said that before, is to bring somebody external to actually train your people and not ask your people to complete courses online just because online courses don't work. Only 3 % of people in the world complete online courses that they're being assigned or paid for, not to mention if their employer assigns it to them and they don't have to pay for the courses themselves. But Microsoft does have a way to win this, and it's actually a relatively easy way, but they have to move fast. And that way is to properly integrate Copilot into everything Microsoft. Right now, what Microsoft has delivered is below par in its results, but it's also not really integrating everything.

47:46You have Copilot for Outlook, you have Copilot for Excel, you have the general Copilot, and they don't necessarily talk to one another. And what they actually need to deliver is a complete integration of Copilot into everything under one umbrella that will allow an employee or a manager or a C-suite person to ask any question and get an answer based on all the company data. That can be SharePoint, Planner, 365 Dynamics, etc., including connections to MCP support, to other tools like Salesforce, Workday, or any other tools that companies use. So a company that is core business run on Microsoft, but has a tech stack with another 30 different SaaS tools, can still see everything within Copilot.

48:20That will be the win that will drive people to use the Microsoft Copilot more than ChatGPT. The problem is ChatGPT is moving in the same direction, right? They're also now allowing you to integrate with Microsoft tools. And so the opportunity window for Microsoft is closing pretty fast. And the only thing that they'll be able to do is provide better integration and better data management and data security. And for right now, it doesn't seem like they're winning this battle. And speaking of AI in enterprises, in ServiceNow's second annual AI maturity index that has surveyed over 4 ,500 worldwide private and public sector leaders, they have found that the average maturity score for AI has dropped from last year.

48:57So from a score of 44 to a score of 35 out of 100 points. In addition, fewer than 1 % of responders scored over 50 in this 100-point maturity scale. And this is obviously very interesting. You would assume that the AI adoption maturity is going to move forward. But the reality is what many companies are seeing, that going from zero to one is relatively easy and getting to a demo stage where the AI does something is not that difficult. But going from there to a production-level product that is consistently generating the right results is not that easy. And hence, many companies are actually moving backwards because now they understand the actual implications and the resources and the limitations that exist that they may have not known a year ago when they started the process and they thought they're further ahead than they actually are.

49:42But this research shares four things that a company should do in order to do this process successfully. The first one is dealing with and managing expectations. And there's two aspects to that. The first one is on the positive side, where these tools can do a lot more than traditional automation tools and even RPA tools because of their ability to actually understand the context and the information in them. So there's a benefit from that perspective of expectation. But on the other hand, a business, especially a large business, is a very complex entity. And combining the entire tech stack and all the different siloed data sources is a lot more complicated than it may seem in the beginning.

50:15and on this topic of smart automation and how not straightforward it is, I have a software company that is developing a solution to automate invoices entry into ERP and accounting systems based on AI and it's the same kind of thing. When we started, we thought it's relatively straightforward and what we've learned over time is that every company is a little different and they have a slightly different tech stack and the invoices come in different ways and they come from different sources and they are saved in different locations and they need to be entered in different ways and every company has a slightly different operation and the ability to have AI support in this process takes it a huge, huge step forward from existing solutions that are either built into ERPs or that comes from traditional RPA software.

50:52So there's definitely a niche in the future for well-integrated AI-based solutions. But as I mentioned, if you're a large enterprise or even a small company and you're working in that direction, you need to be ready for a lot of hurdles and a lot of connections and a lot of integrations and a lot of data security issues that you have to work through. The second topic is pursuing a multi-agent approach. And what they shared that in the research that companies that are successfully implementing this have 10 plus sophisticated agents for different things in the company. So this is, I think, very obvious, but yet it's worth mentioning.

51:20Trying to build a big agent that does a lot of things is a lot harder, a lot more complicated, and the chances of making it successful are significantly lower. Building well-customized tools that do something very, very specific is the way to go, but that means you will need multiple agents in order to actually make an impact in a company, even if it's one single aspect of the business. Let's say supply chain management, you won't build a supply chain management agent. You will build three, four, five different agents that do different things in the supply agent process and will coordinate between them.

51:47The cool thing today is that you can build a multi-layer agent approach where there's one orchestrator or if you want CEO kind of agent that manages the agents underneath that. We're actually recording an episode about this on how to build it yourself, how to build a multi-layer agent environment. And we're going to be releasing it in the next few weeks, but this is definitely the right approach forward when it comes to building agents for companies. The next topic that they're suggesting to think about is rethinking architecture. And they're claiming that most companies are starting with AI agents offered by their existing suppliers, such as their ERP and CRM vendors, because they are natively integrated with the tools they're already using.

52:22That means it's relatively easy to integrate. There's not a lot of work. The disadvantage is that it's limited to just this one tool. And there's a growing interest in purpose-built cross-tool agents that more and more companies are trying to build in order to get this additional value. One of the things I'm teaching in the AI Business Transformation course is how to perform data analysis with data coming from several siloed data sources. So think about crossing your data from your CRM with data from your customer service platform, or data from your supply chain with data from sales, and so on and so forth.

52:52This is stuff that we don't have today. Many big BI tools are built in order to try to provide that, and now you can do it without those fancy tools and without any coders or data scientists, just by knowing how to use AI properly, you can start by manually dragging and dropping CSV files once a day or once a week, whenever you want to get a report. And you can very easily move from that to automation that will do this on its own and will give you up-to-date information all the time without any fancy tools. So extend that into the agent world and having an agent that knows how to connect the dots between multiple aspects of the business is extremely valuable.

53:22And in many cases, more valuable than the agents built into your ERP or CRM because they enable you to look at your business in a way more holistic way than just through this one lens. And the fourth point is refocusing on innovation. In this particular case, what they mean is don't just look for efficiencies. Most companies start with ways to save time for their employees, but the reality is you can dramatically grow your business if you take the right initiatives with AI, whether it's generating leads, nurturing leads, developing new aspects of your product that will drive more revenue, et cetera, et cetera.

53:52And again, going back to the course, the last segment of the AI Business Transformation course is all about strategy. How do you think through the AI lens in order to build better strategy that will harness AI to grow your business and not just to gain daily efficiencies? And then the last item today, as I promised, is going to be about two breakthroughs in the AI space. One comes from Duke University and another from the universities of Surrey and Hamburg that show significant strides forward in AI development. So the Duke study showcases that AI models that mimic human like learning. So they adapt to new tasks without extensive retraining on these new tasks.

54:27This is obviously something that is key to go to AGI, where the G in AGI is general, right? The idea is that you don't have to teach the AI every single specific task, but it will be able to use its knowledge and experience from one task to do another relatively well. And this is exactly what they were able to develop and identify that AI is starting to show the ability to do. Now, the collaboration between the Surrey and Hamburg universities showed something very similar, which is the ability of the AI to reason and problem solve, tackling complex tasks beyond its training capability that is following similar patterns to human cognition.

55:00So what does that tell us? It tells us that the technology, despite the fact its limitations right now of being able to solve problems only in the fields that it was actually trained on, it is moving beyond that. Are we there yet? No, but if you ask all the main leaders, we will be there, meaning a real AGI where AI can actually solve problems beyond its training data is coming in the next one, two, three years. And this basically changes everything we know about AI because at that point, the AI will be able to do more or less everything that we can do in a very effective way across multiple fields, multiple industries, and multiple roles.

55:31And so whenever I hear people saying, oh, it's not gonna take this, it's not gonna impact that, I think we don't really fully understand what this means to have an AI that can really think and reason across every problem while having superpowers of being able to do it very, very fast and have access to basically unlimited data. This is exciting from a technological perspective. It's really, really scary, I think, from a job loss perspective and a social perspective. But if one thing is obvious is that it's not slowing down and that we'll have to figure out how to integrate AI into everything that we're doing.

56:01And hopefully we'll be able to together figure out how to do this in an effective way, enjoy the benefits and avoid as many of the downsides as possible. That is it for today. We will be back with another fascinating how-to episode on Tuesday. go and check out the course. There's a link in the show notes so you can do this right now from your phone or your computer or whatever device you're using to listen. And if you're already there, I would really appreciate it if you click the share button and share this podcast with other people that can benefit from it. This is your contribution to AI literacy for more people.

56:33I'll see you again on the Tuesday episode and have an amazing rest of your week.